A functional neuroimaging study of expectancy effects on pain response in patients with knee osteoarthritis

Abstract

Placebo treatments and healing rituals share much in common, such as the effects of expectancy, and have been used since the beginning of human history to treat pain. Previous mechanistic neuroimaging studies investigating the effects of expectancy on placebo analgesia have used young, healthy volunteers. Using functional magnetic resonance imaging (fMRI), we aimed to investigate the neural mechanisms by which expectancy evokes analgesia in older adults living with a chronic pain disorder and determine whether there are interactions with active treatment. In this fMRI study, we investigated the brain networks underlying expectancy in participants with chronic pain due to knee osteoarthritis (OA) after verum (genuine) and sham electroacupuncture (EA) treatment before and after experiencing calibrated experimental heat pain using a well-tested expectancy manipulation model. We found that expectancy significantly and similarly modulates the pain experience in knee OA patients in both verum (n=21, 11 female; mean ± SD age 57±7 years) and sham (n=22, 15 female; mean ± SD age 59±7 years) acupuncture treatment groups. However, there were different patterns of changes in fMRI indices of brain activity associated with verum and sham treatment modalities specifically in the lateral prefrontal cortex. We also found that continuous EA in knee OA patients can evoke significant regional coherence decreases in pain associated brain regions. Our results suggest that expectancy modulates the experience of pain in knee OA patients but may work through different pathways depending on the treatment modality and, we speculate, on pathophysiological states of the participants.

Introduction

Placebo treatments and healing rituals have been used since the beginning of human history. Although it is well known that expectation of treatment benefit can significantly influence health outcomes⁶³, the systematic study of placebo effects, including expectancy, in patient populations is just beginning.

Of all expectancy effects, analgesia evoked by manipulation of expectation for pain relief is the most thoroughly studied. Neuroimaging studies reveal that activation in complex networks of brain regions is associated with placebo analgesia and the pattern of activation depends on the aspect of the experience being investigated (e.g. during anticipation, pain administration, or post-stimulus pain rating)^{1, 4, 5, 9, 42, 63} and may also depend on the type of stimulus and the method and site of its administration. Most of these studies have been performed on young healthy volunteers; only a few studies^{31, 49, 54, 62} have examined patients with a chronic pain disorder.

Accumulating evidence suggests that expectancy mediated placebo analgesia may work through different brain mechanisms in healthy subjects and patient populations⁴⁹. For instance, one prevailing hypothesis of expectancy-evoked analgesia is that it works through activating the descending pain modulatory system that could subsequently lead to increase in the endogenous opioids that would suppress pain^{18, 68, 77}. However, this system may be altered in chronic pain patients^{34, 55, 73} who are reported to have significant structural changes in key brain regions, such as dorsal lateral prefrontal cortex (DLPFC), and brainstem nuclei^{51, 53} which are activated during pain modulation^{13, 17}. These findings suggest that different mechanisms may mediate expectancy analgesia in people with and without chronic pain disorders.

Much of the research on analgesic expectancy effects investigates only inert treatments such as placebo creams or sham acupuncture. This contrasts with the situation in clinical practice, where expectancy effects interact with an active treatment. Few neuroimaging studies^{6, 12, 41, 43, 57} have investigated expectancy of active analgesic treatments; even fewer studies^{41, 43, 57} have directly compared the brain networks associated with expectancy effects when inert treatments are given to the brain networks associated with expectancy effects when active treatments are given. Because the underlying networks associated with expectancy effects linked with active treatment may differ from those associated with expectancy effects combined with inert treatment, it is imperative to directly investigate the interaction between “pure” expectancy mediated analgesia and treatment mediated analgesia.

Acupuncture is known to produce analgesic effects²⁹ that are mediated, at least in part, through the descending pain modulatory system²⁹. It is also clear that the non-specific effects in acupuncture treatment of chronic pain are robust⁶⁵. Thus, acupuncture treatment in patients with knee OA provides a unique model for investigating the neurobiology of expectancy effects of both verum (genuine) and sham treatments in people living with chronic pain.

In this study, we used multiple analytic methods applied to fMRI data to explore the central brain mechanisms by which expectation modulates the analgesic effects of verum and sham acupuncture treatment in chronic pain patients with knee OA. We hypothesize that different mechanisms underlie expectancy modulation in real versus sham acupuncture in knee OA patients.

Patients and Methods

Participants

All participants were initially screened by means of a detailed telephone questionnaire administered by an experienced research assistant who collected self-report information about current and past medical history, medications, and treatments. All potential participants who passed this initial screen were required to provide written documentation of a recent knee x-ray including the radiology report. The screening information was reviewed by one of our board certified pain medicine clinicians (ADW or RRE) who contacted the participant directly to verify or clarify any inconclusive information. Participants (n=67) were enrolled in the study if they met all inclusion criteria and no exclusion criteria.

Inclusion criteria were: on-going painful knee symptoms in one or both knees; an average daily pain score of at least 3/10 during at least half of the past month; and a grade 2 or 3 on the Kellgren-Lawrence Scale, a radiographic scale suggested by the American College of Rheumatology^{7, 24, 35, 36} for quantifying knee OA severity.

Exclusion criteria were: interventional procedure for knee pain within 6 months, including corticosteroid injections to the knee; intent to undergo surgery during the time of involvement in the study; presence of cardiovascular, neurological or psychiatric disorder that would interfere with conduct of the study; presence of additional pain disorder (including postoperative pain) with severity greater than knee OA pain; pregnancy; positive urine screening test for drugs of abuse; difficulties in reading, speaking or understanding English; or contraindications to MRI scanning. Further, during the analysis stage, the participants were excluded if they did not complete all of the study sessions; showed excessive head movement during MRI scan; or had a brain abnormality discovered during the MRI scan.

All participants were naïve to acupuncture at the time of enrollment so that they could not guess if they were receiving real or sham acupuncture and to also avoid confounds due to previous experience or conditioning effects. The participants were told that this study was an investigation of acupuncture analgesia. Experiments were conducted with approval from the MGH IRB and written, informed consent of each participant (ClinicalTrials.gov identifier: NCT01040754). Participants were debriefed about the true nature of the study after completion of all procedures. All participants agreed to allow their data to be analyzed.

Experimental procedures

In a previous manuscript we reported on the feasibility of using pre-test resting state fMRI data from this experiment to predict treatment response to verum and sham EA³². The present study focuses on analyses of fMRI scans which have not been reported before. The experimental design was nearly identical to our previous studies^{41, 43} conducted in healthy subjects. Each patient participated in two behavioural testing sessions the first of which included neuropsychological assessments including the Beck Depression Inventory (BDI)²¹ and Spielberger State-Trait Anxiety Inventory (STAI)⁶⁰, and one fMRI scanning session. Each session was separated by a minimum of three days from the other sessions with no more than ten days of separation between consecutive sessions.

Calibrated thermal pain stimuli were delivered to the right medial aspect of the forearm using a TSA-2001 Thermal Sensory Analyzer with a 3 cm × 3 cm probe (Medoc Advanced Medical Systems, Rimat Yishai, Israel) running computerized visual analog scale software (COVAS). All stimuli were initiated from a baseline resting temperature of 32 °C and increased to a target temperature determined according to each participant’s sensory and affective ratings for nociceptive stimulation (see below). Each stimulus was presented for 12 seconds, including a 2.5 second ramp up and ramp down, and the inter-stimulus interval ranged from 24–30 seconds.

Gracely Sensory and Affective scales^{2, 22, 25–28, 50} were used to measure subjective pain ratings. To ensure consistent pain administration, a 2×3 grid was drawn with marker along the palmar side of the forearm, with three boxes each on radial and ulnar sides. We placed the thermal probe in one box of the grid for each stimulus sequence (Figure 1).

Figure 1. Study design and experimental approach.

Enrolled participants are trained to rate calibrated, exogenous heat pain stimuli in Session 1. In Session 2, participants that rate the session 1 training stimuli consistently across these two sessions proceed in the study. They are randomized to either sham or verum treatment and to be treated on either radial or ulnar side of their right forearm. The expectancy manipulation consists of verbal suggestion that acupuncture will diminish heat pain paired with a conditioning procedure only on the “treated” side. The conditioning consists of administration of random series (RS) and identical series (IS) of noxious heat stimuli to each of the three regions on each side of their right forearm (PRE), applying either sham or verum EA to their right hand at LI3 and LI4 acupoints, and then repeating the IS of noxious heat stimuli (POST) with temperatures surreptitiously lowered on the high expectancy side (dIS in green) to create the unmistakable experience of pain relief. The neuroimaging experiment is conducted in Session 3. Participants are told that the Session 2 procedures will be repeated during fMRI scans. To further boost expectation of pain relief, the first series of noxious heat stimuli on the treated side are once again delivered at a lower temperature. The fMRI scans collected during four series of identical noxious stimuli to the “treated” and control sides of the forearm are the outcome measures.

Session 1

We used the first session to determine appropriate heat stimulus intensities for each participant; to minimize anticipatory anxiety; to control for rating strategy and learning effects; and to teach the participants to rate noxious heat stimuli using the Gracely Sensory Box and Affective Box scales²⁷. The Box Scales are ratio scales used to rate perception of the sensory and affective components of pain sensations that are particularly sensitive to determining change within an experimental session.

Supra-threshold heat pain stimuli were delivered to the right medial forearm. The thermode was placed over one of the six regions and left in place for each sequence of stimuli. Two temperatures, one that elicited LOW ratings (5–7 on the Gracely Scale; mild to moderate) and one that elicited HIGH ratings (14–17; strong to intense) were selected for each participant.

Session 2

The pain stimulation during this session consisted of four phases. 1) Administration of a random sequence (RS) consisting of eight randomized HIGH and LOW temperature stimuli. This RS sequence was administered twice, i.e. once to each of the bottom two boxes on the grid. 2) Administration of an identical sequence (IS) consisting of six repetitions of the same HIGH intensity stimulus four times, i.e. once to each of the top four boxes of the grid. 3) Treatment with either verum or sham EA. And 4) post-treatment expectancy manipulation described below (Figure 1).

Participants had to consistently rate HIGH pain stimuli as more painful than LOW pain stimuli during the RS administration, report approximately equivalent pain intensity ratings to the IS on the radial and ulnar sides of their arm, and also be consistent in their ratings to these two sequences from Session 1 to Session 2. If so, they were randomized into one of the treatment groups and continued in the study; otherwise they were dropped.

Expectancy Manipulation

We manipulated the participants’ expectancy of acupuncture analgesia using the same method as employed in our previous studies in healthy subjects^{40, 41, 43} as modified from earlier studies using placebo creams⁶⁷.

At the beginning of Session 2 participants were given a scripted explanation that a person’s responses to acupuncture can be positive or neutral, and that this response tends to remain consistent over time. Participants then viewed a traditional Chinese medicine meridian diagram and were told that acupuncture could only produce analgesia on the side of the arm through which the meridian passed (the “treated” side, where the needles would be placed) but not on the other side of the arm (the “untreated” side). Neither of these statements is in fact true. This verbal suggestion is one component of the manipulation to enhance expectancy for analgesia on the treated side. To balance the design, half the participants were shown accurate diagrams of the Large Intestine meridian passing through the radial side of the arm, while the other half viewed a modified diagram showing the meridian passing through the ulnar side of the arm. The conditioning aspect of the expectancy manipulation consisted of telling participants that they would receive identical heat pain stimuli before and after treatment; but in reality, after treatment we surreptitiously lowered temperatures on the “treated side” (high expectancy, HE side) of the arm sufficiently to reduce their rating to “faint to weak” and hence give participants an unmistakable experience of profound analgesia. On the “untreated” (Control) side, the temperatures remained at pre-treatment HIGH levels to enhance the “good effect” of acupuncture treatment.

Acupuncture administration

Verum or sham acupuncture was performed at Large Intestine 3 and 4 on the right hand by a licensed acupuncturist. For verum EA, needles were adjusted until deqi, an unique sensation that indicates efficacious acupuncture³⁸, but no sharp pain, was evoked. Needles were then connected to an electroacupuncture device passing a continuous 2Hz current (OMS Medical Supplies IC-1107)^{37, 41, 43}. For sham EA, validated Streiberger sham acupuncture needles were placed on the surface of the skin and connected to a de-activated EA device^{40, 61, 76}. These placebo needles retract into the barrel when pressed on the skin, similar to the action of a retractable stage dagger. Verum or sham EA treatment lasted approximately 25 minutes.

Session 3

Session 3 was performed in a 3T MRI scanner. Participants were told we would repeat the same procedures as in Session 2. Participants received same verum or sham EA on the same “treated” side as they had in Session 2. After treatment, participants were told that we would again present the identical pain stimuli. As in Session 2, we decreased the post- treatment stimulus temperature to LOW intensity on the “treated”, HE side, but only on the first sequence to further reinforce or boost their expectancy for reduced pain on the “treated” side. Then, we measured the participants’ expectancy to the analgesic effect of the acupuncture treatment using a 0–10 scale (0 does not work at all, 10 completely relieves pain). Finally, the original HIGH intensity IS was delivered to each of the remaining four boxes. BOLD fMRI sequences were run during all applications of pain stimuli as well as during the verum or sham EA treatment. Figure 1 shows a diagram of the experimental design and schematic of the procedures in each session

fMRI data acquisition and data analysis

Whole brain imaging during delivery of each of the RS and IS sequences of pain stimuli as well as during verum or sham EA treatment was performed with a 12-channel head coil in a 3T Siemens TIM Trio MRI System. BOLD images were collected with thirty axial slices (4mm thick with 1mm skip) parallel to the anterior and posterior commissure, 2000ms TR, 40ms TE, 90° flip angle and 3.13 × 3.13mm in-plane spatial resolution. A 3D MPRAGE sequence was used for high-resolution anatomical data acquisition.

Data were analyzed with SPM8 software. Motion correction, co-registration, normalization and smoothing (8mm) were implemented in SPM8. Outliers in the fMRI time series that were potentially due to participants’ motion were identified using the ART software^{10, 69}. Images with mean signal intensity three standard deviations or more beyond the mean and with head movement exceeding 0.5mm or 0.015 radians for translational and rotational movement respectively (relative to the previous scan) were identified and regressed out in the statistical analysis.

For each individual, the fMRI signal difference between HIGH pain and LOW pain conditions during the two RS scans, and the fMRI signal difference between identical HIGH pain stimuli pre- and post-treatment during the two IS scans on the HE side and on the Control side were calculated for each participant using a general linear model. Group analysis was performed using a random-effects model. For this and all other fMRI analyses, we included age as a covariate of no interest. The contrast of all pre-treatment HIGH pain with all LOW pain stimuli on both HE and Control sides from the RS sequences^{40, 67} was used to generate a mask of pain-intensity encoding brain regions with which we could test for the treatment mode and expectancy effects in the subsequent analyses. A voxel-wise threshold of p<0.005 with p<0.05 cluster level false discovery rate (FDR) correction was applied.

To investigate the effects produced by treatment mode and expectancy, we performed the following analyses: 1) compared pre- and post- fMRI signal change differences between the HE and Control sides within each of the verum and sham acupuncture groups using a paired t-test; 2) compared treatment modes (verum and sham) indicated by pre- minus post-treatment differences in the HE and Control sides separately using a two sample t-test; 3) estimated the interaction between treatment modes (verum versus sham). This interaction was calculated by comparing fMRI signal changes during pain administration between verum acupuncture on HE side and sham acupuncture on the Control side, and verum acupuncture on the Control side and sham acupuncture on the HE side (verum ((HE (pre – post) – Control (pre- post)) – Sham ((HE (pre – post) – Control (pre- post))). Based on ours and others previous imaging studies of pain, placebo analgesia and pain modulation^{3, 6, 18, 40, 42, 57, 67}, our a priori regions of interest included: 1) pain-intensity encoding brain regions from the RS mask (see above) and 2) pain modulation network- i.e. those brain regions reported to be involved in pain modulation and cognition including orbital prefrontal cortex (OPFC), medial prefrontal cortex(MPFC)/ACC, and lateral prefrontal cortex (LPFC) defined based on AAL (http://www.fil.ion.ucl.ac.uk/spm/ext/#AAL). An initial threshold of p<0.005 was used in all data analysis. To correct for multiple comparisons, Monte Carlo simulations using the 3dFWHMx and 3dClustSim (as part of the AFNI program (https://afni.nimh.nih.gov) released in July 2017) were applied for the ROIs (i.e. pain encoding and modulation regions listed above, where for each region, the minimum voxel size required for p < 0.05 cluster level p-value correction will be indicated as k value in the results presented below). For the rest of the brain, a voxel-wise threshold of p<0.005 with p<0.05 cluster level FDR corrected was applied.

Regional functional connectivity analysis

In order to acquire further insights into the local changes in the brain connectivity associated with different treatment modalities, we used a data-driven, resting state functional connectivity analysis of the regional homogeneity (ReHo)^{33, 74, 75}, to investigate regional coherence differences between verum and sham EA in the fMRI data acquired during treatment. ReHo uses Kendall’s coefficient of concordance to measure the similarity of time courses of voxels within a given cluster. The method assumes that the within a functional cluster, time courses of the fMRI signal from the individual voxels within the cluster are similar to each other, reflecting the temporal synchrony of the regional BOLD signal.

All pre-processing steps and the ReHo analysis were performed using Data Processing Assistant for Resting State fMRI (DPARSF) Advanced edition based on SPM8¹⁴. Pre-processing included: removing the first 10 time points, slice timing correction using the middle slice as a reference, realignment (motion correction), normalization into MNI space using an EPI template, linear trend removal, and filtering results to those in the range of 0.01–0.08Hz. Next, the regional homogeneity was calculated using a 27 voxel cluster. At a given voxel, ReHo was defined as the Kendall’s Coefficient of Concordance of the time series of this voxel with those of its 26 nearest neighbors. The resulting map of ReHo values at each voxel was then divided by the global mean value within the whole-brain mask. This result was then smoothed with a 6mm kernel. Finally, a t-test was performed in SPM8 to compare the difference in ReHo values at each voxel between groups receiving verum and sham EA treatment. A voxel-wise threshold of p<0.005 with p<0.05 cluster level FDR correction was applied.

Results

Of the 67 participants who participated in the study, 45 (26 female, average age = 58.7±7.4, 22 in EA group) completed the study. Three participants were dropped for failing the drug abuse screen in the first session, eight for inconsistent pain ratings in the second session and one due to a previously unknown, pre-existing brain abnormality. One participant withdrew after a power outage in the scanner halfway through the experiment interrupted the study, one because she found the acupuncture sensations to be too uncomfortable, and four because they did not feel comfortable enough in the MRI machine to complete the study procedures. Four participants did not continue after the first session due to scheduling difficulties.

Of the 45 participants who completed the study, fMRI data from additional two participants were dropped, one due to excessive head movement during scan, another due to a pre-existing brain abnormality that was not discovered during the scan visit. Thus, a total of 43 participants, n=21 verum EA (11 female), age 57±7 years (average±SD) and n=22 sham EA (15 female), age 59±7 years were included in the final data analysis. No significant differences in age or gender distribution were found between the two groups. The average temperature used for the low pain stimuli was 45.7 ± 1.7 °C for the verum group and 45.6±1.0 °C for the sham group. For the high pain stimuli, the average temperature used was 48.1 ± 1.4 °C in the verum group and 48.0 ± 0.8 °C in the sham group.

We did not find evidence of major psychiatric disorders in any participants. BDI and STAI scores were within normal range (BDI 5.2±6.7; STAI 25.2±5.9). None of these assessments differed between the two groups (verum and sham) or were significantly correlated with the expectancy mediated analgesic effect.

Subjective Ratings of Pain and Expectancy

Pain reduction following verum or sham EA was calculated by subtracting the pain rating for each post-manipulation trial from the pain rating for each pre-manipulation trial and averaging across trials in the sequences. Means and standard deviations of pain ratings to sequences of identical heat pain stimuli are displayed in Table 1. These data were analyzed using a 2×2 (told by received) mixed model analysis of variance (ANOVA), in which what participants were told (HE versus Control side) was a within-subject factor, and what they received (verum or sham EA) was a between subject factor.

Table 1.

Subjective pain ratings (mean ± SD) to sequences of identical heat pain stimuli (IS) in verum and sham EA groups. HE, High Expectancy.

	Verum EA		Sham EA
	Pre-treatment	Post-treatment	Pre-treatment	Post-treatment
HE side	12.5 ± 2.2	11.4± 2.2	13.4 ± 2.4	12.2 ± 3.3
Control side	12.4 ± 2.0	13.0 ± 2.3	13.5± 2.4	14.0 ± 2.4

The ANOVA revealed a significant main effect on what participants were told (i.e. the effect of manipulation of expectancy), F (1,41) = 18.417, p<0.001 in both verum and sham groups. Participants reported greater pain reduction on the HE side than on the Control side of their forearm (Table 1, Figure 2). There were no other significant main effects or interactions. Treatment modality effect (verum versus sham EA) was not significant, F (1,41) =0.558, p=0.459, nor did the interaction between modality and what participants were told approached significance, F (1,41) =0.010, p=0.920. Within-subject t-tests revealed that both verum and sham EA decreased pain significantly on the HE side of the forearm (p<0.001), and further that participants experienced increased pain on the Control side (p<0.001).

Figure 2.

Scatter plot showing change in each individual participant’s pain rating on the 0–20 Gracely Scale from pre- to post-treatment on the high expectancy (HE) and control sides of their forearm for both the verum and sham groups. Each rating shown is the average of 12 identical moderate heat pain stimuli. ANOVA revealed a significant main effect on what participants were told (i.e. the effect of manipulation of expectancy), F (1,41) = 18.417, p<0.001 in both verum and sham groups. Treatment modality effect (verum versus sham EA) was not significant, F (1,41) =0.558, p=0.459, nor did the interaction between modality and what participants were told approached significance, F (1,41) =0.010, p=0.920. Both verum and sham EA decreased pain significantly on the HE side of the forearm (p<0.001), and participants experienced increased pain on the Control side (p<0.001).

To put into perspective the behavioural expression of placebo effects in the older chronic pain population of this study (average age = 58.7±7.4) as compared to younger healthy subjects (average age 26.4 ± 4.9), we also directly compared the results of this study with our previous study^{41, 43}. We added type of participants to the ANOVA, resulting in a 2×2×2 (participant status by told by received) mixed model analysis of variance (ANOVA), in which what participants were told (HE versus Control side) was a within-subject factor, and what they received (verum or sham EA) and participant status (healthy or chronic pain) were between subject factors. This ANOVA revealed a significant main effect on what participants were told, F (1,63) = 32.693, p<0.001 with participants reporting greater pain reduction on the HE side than on the Control side. There were no other significant main effects or interactions, suggesting the magnitude and direction of expectancy effects as indicated by subjective pain rating changes in older chronic pain patients are comparable to that observed in young healthy subjects. We note that age could not be added as a covariate in this analysis, because it is perfectly correlated with group. Had we obtained a group difference, this confound would have posed an interpretive problem. However, since we did not, our results make it clear that neither age nor patient status make a difference in the conditioning effect we obtain.

In this study, expectation of pain relief measured before the application of post-treatment IS sequence was 7.6±1.3 for the verum EA group and 7.5±1.5 for the sham EA group, indicating that all participants had a strong belief that treatment would be effective in reducing their pain and no difference in expectation of pain relief between the verum and sham EA groups. We performed additional analyses to investigate relations between pain scores at baseline, expectancy ratings both at baseline and during the scan, and pre-post ratings on the HE side of the arm. We found no significant correlations between measures of knee pain at time of scanning and expectancy effects in either the verum or sham acupuncture groups.

At the conclusion of the study, all participants completed a final questionnaire using a scripted assessment on a computer program inquiring about whether they thought a needle was inserted into their skin in each session, whether they thought any of the sessions were a placebo treatment, and their certainty of their answers. Consistent with previous studies using the Streitberger needles for sham acupuncture, 19 of the 21 participants in the sham cohort reported with great confidence that they thought real needles were inserted, as did all the participants in the verum cohort. Six of the 21 participants in the sham group believed that at some point in the study they received a placebo treatment, while only 2 of the 22 that received verum acupuncture thought that.

fMRI results

Evoked pain

The contrast between all pre-treatment HIGH and LOW pain stimuli during the RS sequence yielded significant activations in the entire predicted network of pain intensity sensitive regions, including bilateral insula, dorsal anterior cingulate cortex/medial prefrontal cortex (dACC/MPFC), secondary somatosensory cortex (S2), thalamus, putamen, periaquaductal grey (PAG), cerebellum, precuneus, and left (contralateral) SI/M1 (Figure 3)^{3, 6, 39, 40, 44, 47}.

Figure 3. Evoked pain activation maps.

The group average statistical map of the fMRI signal change difference evoked by HIGH vs. LOW heat pain intensity stimuli during the initial random sequence (RS) are presented. The group average across all participants of the study is shown in the first row. This pain network activation map was used as the mask for subsequent analysis.

Effect of expectancy (within group comparisons)

The results of the within group HE vs. Control side comparisons (HE side (pre-treatment minus post-treatment) versus Control side (pre-treatment minus post-treatment)) are shown in Table 2 and Figure 4. In the verum EA group (Figure 4A), identical noxious stimuli delivered to the HE side produced greater signal increases after treatment than those delivered to the Control side (HE side minus Control side) in left (contralateral to pain stimuli) LPFC (k = 32), right lateral orbital prefrontal cortex (LOPFC) (k = 32), and right insula/putamen (k = 83). There was no brain region that showed a significant difference for the opposite comparison.

Table 2.

Comparison results of fMRI signal change differences (pre-treatment pain > post-treatment pain) between High Expectancy (HE) and Control sides of the forearm in response to identical heat pain stimuli, and regression analysis between the subjective pain rating change and corresponding fMRI signal changes in verum and sham EA treatment groups.

Comparisons		Brain Region	Z score	Number of voxels in cluster	Peak coordinate (x,y,z)
Expectancy effect with verum EA	HE > Control	No region above threshold
	Control > HE	Right lateral orbital prefrontal cortex	3.44	162	−52 40 −4
		Left lateral prefrontal cortex	3.17	113	52 12 14
		Right insula / putamen	3.20	145	−32 8 2
Expectancy effect with sham EA	HE > Control	No region above the threshold
	Control > HE	Bilateral subgenual ACC / MPFC	3.73	100	2 44 −16
		Right lateral orbital prefrontal gyrus	3.71	116	−36 46 −12
Regression analysis in verum EA group	Positive	No region above threshold
	Negative	No region above the threshold
Regression analysis in sham EA group	Positive	Left posterior medial prefrontal cortex	3.6	162	14 42 36
		Left occipital cortex	3.43	761	14 −78 6
		Right occipital cortex	3.56	971	−20 −74 34
	Negative	No region above threshold

Figure 4. Effect of expectancy on brain activation in verum vs. sham EA.

The results of the within group HE vs. Control side comparisons (HE (pre-treatment minus post-treatment) minus Control side (pre-treatment minus post-treatment)) are shown in this figure. The brain networks involved in expectancy effect in the group treated with verum EA (A), in the group treated with sham EA (B), and the difference between the two (C) are shown. The activation identified in the insula/putamen region (3A and 3C) is found within the pain-intensity encoding mask generated by the High-Low pain contrast in the RS scans (shown in Figure 3). The other regions of activation found in the LPFC, OPFC and sACC/MPFC (4A and 4B) are found within the mask of the pain modulation network derived from the literature^{3, 6, 18, 40, 42, 57, 67}. sACC, sub-genual anterior cingulate cortex; LPFC, lateral prefrontal cortex; MPFC, medial prefrontal cortex; OPFC, orbital prefrontal cortex. L, left side.

In the sham EA group, identical noxious stimuli delivered to the HE side produced more signal increases after treatment than those delivered to the Control side in bilateral sub-genual ACC (sACC)/MPFC (k = 75), and right OPFC (k = 32) (Table 2 and Figure 4B). There was no brain region that showed a significant difference for the opposite comparison.

Differences in the effect of expectancy between verum and sham groups

To directly compare the effect of expectation in the verum and sham EA groups, we performed a two sample t-test restricted to the five regions (bilateral sACC/MPFC, left LPFC, right OPFC, insula/putamen) that showed a significant pre-post treatment difference between the HE and Control sides in either of the two groups. We found no significant differences between the two groups at the threshold we set; however, we did find a significant difference in the right insula/putamen (peak coordinate (x, y, z): −28, 10, −4, 109 voxels, k = 84) with a threshold of voxel-wise p<0.05 uncorrected and a corrected threshold of p<0.05 at the cluster level based on Monte Carlo simulations with the 3dFWHMx and 3dClustSim (see the methods section) (Table 2 and Figure 4C). There was no brain region that showed a significant difference when we applied a whole brain correction.

Effect of verum vs. sham EA treatment

To investigate the effects of acupuncture treatment, we compared pre- and post-treatment fMRI signal differences during pain stimulation between verum and sham EA on HE and Control sides separately. We found no significant difference between verum and sham. To explore the relation between brain activity change and subjective pain rating change, we also performed a regression analysis (Table 2). A significant positive association was observed in the right MPFC (k = 89), and bilateral occipital cortex in the sham EA group. No region was significant in the verum EA group.

Regional functional connectivity difference in verum vs. sham EA

Regional coherence, as measured by the ReHo analysis, increased in left postcentral and precentral gyrus in an area corresponding to the right hand where the EA was applied in the verum group more than the sham group (Table 3, Figure 5). Conversely, compared with sham EA, during verum EA, the regional coherence significantly decreased in bilateral dorsal ACC/MPFC, right S2, supramarginal gyrus/superior temporal gyrus, inferior parietal lobule, and left insula/operculum.

Table 3.

Regional homogeneity difference between verum and sham EA

Comparisons	Brain Region	Peak coordinate (x,y,z)	Z score	Number of voxels in cluster
Verum > sham	Left postcentral/precentral gyrus	52 −19 56	3.86	265
Sham > verum	Bilateral dorsal ACC / medial prefrontal cortex	7 16 28	3.98	483
	Left insula / operculum	36 −10 4	3.87	275
	Right supramarginal gyrus / superior temporal gyrus	−49 −57 28	4.29	233
	Right secondary somatosensory cortex	−43 −32 24	3.65
	Right inferior parietal lobule	−37 −47 52	3.23

Figure 5. Regional coherence difference between EA and sham EA in knee OA patients.

Brain regions that showed significant regional coherence increase during verum EA as compared with sham EA are shown in (A). Brain regions that showed significant regional coherence decrease during verum EA as compared with sham EA are shown in (B). dACC, dorsal anterior cingulate cortex; MPFC, medial prefrontal cortex; S2, secondary somatosensory cortex; L, left side.

To further explore the different brain response during EA in patients and healthy controls, we also re-analyzed our data set in the younger healthy cohort that used a similar paradigm^{41, 43}. We found no significant difference between verum and sham at the statistical threshold we set. At a less conservative threshold of p < 0.05 uncorrected with 20 continuous voxels, we found that compared with sham EA, verum EA produced significant regional coherence increases in bilateral precentral and postcentral gyrus, and left insula.

Discussion

We investigated the effects of verum electroacupuncture and sham acupuncture on heat pain ratings and associated BOLD responses in patients with chronic pain due to knee OA. We also used a well-studied expectancy manipulation model that combines conditioning and verbal suggestion to try to enhance the treatment-induced analgesia. We found that increased expectation of pain relief by our manipulation significantly decreased subjective reports of heat pain intensity in response to calibrated experimental noxious stimulation after either verum or sham EA in knee OA patients. This matched expectancy effect was associated with different brain activity patterns, as measured by fMRI, after verum and sham EA. Our results reflect greater activity after treatment in specific brain regions when responding to identical noxious stimuli applied to areas of skin on the arm where subjects expect to feel less pain (HE side) as compared to the control side due to expectation manipulation. The specific brain regions that showed this greater activity differed for the groups treated with verum and sham acupuncture. Within the verum EA group, stronger activation was observed in LPFC, OPFC and insula/putamen at the time of identical noxious stimulation when there was a high expectation of analgesia. The same contrast in the sham EA treatment group revealed increased activation in sACC/MPFC, and OPFC. In addition, brain activity change in the MPFC and occipital cortex in the sham EA group, but not in the verum EA group, was positively associated with subjective pain rating changes. These results suggest that expectancy effects may be modulated by distinct mechanisms when coupled with different modalities of treatment.

Previous placebo imaging studies, all in healthy subjects, investigating how expectancy can modulate the analgesic effect of active treatments have yielded inconsistent behavioural results. Using a conditioning-like expectancy manipulation similar to that used in our study, Bingel and colleagues found¹² that positive expectancy can significantly enhance the analgesic effect of remifentanil, while negative expectancy can antagonize its analgesic effect. In another study, Atlas and colleagues⁶ using a verbal suggestion balanced placebo paradigm found that both expectancy and remifentanil can reduce pain, with no significant interaction between these two factors. In a more recent verbal suggestion, balanced placebo study using a topical analgesic treatment, Schenk and colleagues⁵⁷ investigated the interaction between lidocaine and expectancy using a clinical pain related model (capsaicin pre-treated skin). They found that lidocaine significantly reduced pain as compared with placebo treatment, while the main expectancy effect (open administration compared with hidden administration) was not significant. However, unlike Atlas et al.⁶, they found a significant interaction between treatment and expectancy. Open administration of lidocaine reduced pain rating significantly more than hidden, while there was no significant difference between open and hidden administration of placebo.

The inconsistency in how expectation of pain relief impacts an active treatment in these experimental settings may be due to a number of factors including differences in expectancy manipulation (verbal suggestion alone versus a more powerful conditioning-like procedure^{16, 45, 67}) and differences in treatment modality (remifentanil, lidocaine, and acupuncture). Taken together, these results suggest that the effect of expectancy on treatment outcome may depend on the specific details of the expectancy manipulation as well as the treatment modality. Due to the experimental design of this study, it is not possible to investigate within subject differences in how expectancy interacts with sham and verum acupuncture.

Our experimental design, which studied the neural activity associated with placebo effects in older patients with a chronic pain disorder, revealed significant fMRI signal increases in brain regions including bilateral sACC and right lateral OPFC after sham EA treatment on the HE side as compared with Control side. Using fMRI and a within subject design, Schmid and colleagues showed that in the setting of comparable expectations of pain relief and comparable placebo analgesia, patients with chronic pain due to irritable bowel syndrome (IBS) activate a similar region of the cingulate cortex in both the placebo and control conditions; while patients with ulcerative colitis in remission and age- and sex- matched healthy controls have placebo related reductions in cingulate cortical activation⁵⁸. However, Price and colleagues showed a large placebo response in the form of reduction of brain activity in the anterior cingulate cortex in IBS patients as well as in the thalamus, somatosensory cortices, and insula⁵⁴. Our results are partly consistent with another previous study⁴⁹ in IBS patients that found increased activity in lateral OPFC associated with placebo analgesia. It is known that the ACC is a component of the descending pain modulation system^{20, 46, 73}. The increased fMRI signal in ACC that we find is consistent with previous placebo analgesia studies observed in healthy subjects^{11, 18, 23, 40, 77}, leaving open the possible interpretation that modulation of the descending pain modulation system is critical for placebo analgesia.

Direct comparison between the verum and sham EA groups for differences in expectancy effects showed significant differences in the right insula/putamen. This result is consistent with the interpretation that there are both common and distinct components of the brain networks that underlie expectancy effects when combined with verum or sham EA. In the verum EA group significant pain evoked fMRI signal increases were observed in right insula/putamen, lateral OPFC and left LPFC when comparing the Control side vs. the HE side whereas in the sham group the effect of expectancy was observed in the sACC/MPFC and OPFC. Although not overlapping, the areas of increased activation in the right lateral OPFC observed in verum and sham EA groups are adjacent to each other, suggesting that the right lateral OPFC is involved in expectancy effect across physiological and pathological states, which supports a crucial role of lateral PFC in top-down modulatory mechanisms of analgesia^{42, 48, 63, 64, 66, 70}.

Functional connectivity analysis showed that regional coherence in the insula, operculum, S2, and dorsal ACC was significantly reduced during verum EA as compared to during sham EA, indicating that the pain network was significantly modulated during EA stimulation. This result was not observed in a cohort of healthy subjects studied in a very similar paradigm^{41, 78}. These results suggest that the whole brain response to verum acupuncture treatment is significantly different in people with chronic pain as compared to healthy controls, which we speculate may be in part responsible for the different expectancy evoked neural networks.

The finding of modulation of the pain network in verum compared to sham in patients with knee OA may be the key to interpreting the difference in expectancy between these two treatment modes. As shown in Figure 4C, the difference in expectancy effect between the verum vs. sham is limited to the putamen/striatum. Based on our ReHo findings, it is likely that the difference of expectancy effect may be driven by the activation of the pain network. The observed activation in the putamen/striatum suggests that the modulation of pain network by verum acupuncture treatment may involve alterations in dopaminergic and opioidergic activity and involve the reward circuits^{15, 30, 52, 71}. This may be due to adaptive changes in the brain as a consequence of ongoing nociceptive input in knee OA^{56, 59}. Further, this difference in response during EA may explain why we did not find any significant fMRI signal decreases in the present study of knee OA patients when we compared verum to sham EA as we did in our previous study in healthy subjects⁴¹.

The patients who participated in this study were significantly older compared to participants in reported studies on healthy subjects. Although aging is associated with substantial functional and structural changes in areas of the brain known to be involved in placebo responses¹⁹, a recent study investigating age-related changes has demonstrated preservation of the placebo response in the elderly using a paradigm similar to the one used in this study⁷². Benedetti and colleagues⁸ found that Alzheimer’s disease (AD) patients showed reduced placebo analgesia one year after the first test. Interestingly, they also found that healthy elderly volunteers had robust placebo analgesic responses, further supporting the finding that psychobiological mechanisms underlying the efficacy of placebo effects are present in cognitively intact older people. While our results support these results and further suggest that these mechanisms remain intact even in the presence of chronic pain due to knee OA, additional work will need to be done regarding generalization to other chronic pain disorders.

In conclusion, our study suggests that expectancy may work through different pathways based on the treatment modality and pathophysiological state of the person. Future studies that include larger cohorts of chronic pain patients and age matched pain-free control subjects that can be directly compared are needed to disentangle the potential confounding effects of aging on these neural responses.

Perspective.

To investigate the neural mechanisms underlying pain modulation, we used an expectancy manipulation model and fMRI to study response to heat pain stimuli before and after verum or sham acupuncture treatment in chronic pain patients. Both relieve pain and each is each associated with a distinct pattern of brain activation.